327 lines
13 KiB
Python
327 lines
13 KiB
Python
from random import random
|
|
from random import gauss
|
|
from agent.Base_Agent import Base_Agent as Agent
|
|
from behaviors.custom.Step.Step import Step
|
|
from world.commons.Draw import Draw
|
|
from stable_baselines3 import PPO
|
|
from stable_baselines3.common.vec_env import SubprocVecEnv
|
|
from scripts.commons.Server import Server
|
|
from scripts.commons.Train_Base import Train_Base
|
|
from time import sleep
|
|
import os, gym
|
|
import numpy as np
|
|
from math_ops.Math_Ops import Math_Ops as U
|
|
from math_ops.Math_Ops import Math_Ops as M
|
|
from behaviors.custom.Step.Step_Generator import Step_Generator
|
|
|
|
'''
|
|
Objective:
|
|
Learn how to run forward using step primitive
|
|
----------
|
|
- class Basic_Run: implements an OpenAI custom gym
|
|
- class Train: implements algorithms to train a new model or test an existing model
|
|
'''
|
|
|
|
|
|
class Kick_15M(gym.Env):
|
|
def __init__(self, ip, server_p, monitor_p, r_type, enable_draw) -> None:
|
|
self.lock_flag = False
|
|
self.sleep = 0
|
|
self.reset_time = None
|
|
self.behavior = None
|
|
self.path_manager = None
|
|
self.bias_dir = None
|
|
self.robot_type = r_type
|
|
self.kick_ori = 0
|
|
self.terminal = False
|
|
# Args: Server IP, Agent Port, Monitor Port, Uniform No., Robot Type, Team Name, Enable Log, Enable Draw
|
|
self.player = Agent(ip, server_p, monitor_p, 1, self.robot_type, "Gym", True, enable_draw)
|
|
self.step_counter = 0 # to limit episode size
|
|
self.ball_pos = np.array([0, 0, 0])
|
|
|
|
self.step_obj: Step = self.player.behavior.get_custom_behavior_object("Step") # Step behavior object
|
|
|
|
# State space
|
|
obs_size = 63
|
|
self.obs = np.zeros(obs_size, np.float32)
|
|
self.observation_space = gym.spaces.Box(low=np.full(obs_size, -np.inf, np.float32),
|
|
high=np.full(obs_size, np.inf, np.float32), dtype=np.float32)
|
|
|
|
# Action space
|
|
MAX = np.finfo(np.float32).max
|
|
self.no_of_actions = act_size = 16
|
|
self.action_space = gym.spaces.Box(low=np.full(act_size, -MAX, np.float32),
|
|
high=np.full(act_size, MAX, np.float32), dtype=np.float32)
|
|
|
|
# Place ball far away to keep landmarks in FoV (head follows ball while using Step behavior)
|
|
self.player.scom.unofficial_move_ball((14, 0, 0.042))
|
|
|
|
self.ball_x_center = 0.21
|
|
self.ball_y_center = -0.045
|
|
self.player.scom.unofficial_set_play_mode("PlayOn")
|
|
self.player.scom.unofficial_move_ball((0, 0, 0))
|
|
|
|
def observe(self, init=False):
|
|
w = self.player.world
|
|
r = self.player.world.robot
|
|
|
|
if init:
|
|
self.step_counter = 0
|
|
self.obs[0] = self.step_counter / 20
|
|
self.obs[1] = r.loc_head_z * 3
|
|
self.obs[2] = r.loc_head_z_vel / 2
|
|
self.obs[3] = r.imu_torso_roll / 15
|
|
self.obs[4] = r.imu_torso_pitch / 15
|
|
self.obs[5:8] = r.gyro / 100
|
|
self.obs[8:11] = r.acc / 10
|
|
self.obs[11:17] = r.frp.get('lf', np.zeros(6)) * (10, 10, 10, 0.01, 0.01, 0.01)
|
|
self.obs[17:23] = r.frp.get('rf', np.zeros(6)) * (10, 10, 10, 0.01, 0.01, 0.01)
|
|
self.obs[23:39] = r.joints_position[2:18] / 100
|
|
self.obs[39:55] = r.joints_speed[2:18] / 6.1395
|
|
ball_rel_hip_center = self.player.inv_kinematics.torso_to_hip_transform(w.ball_rel_torso_cart_pos)
|
|
if init:
|
|
self.obs[55:58] = (0, 0, 0)
|
|
elif w.ball_is_visible:
|
|
self.obs[55:58] = (ball_rel_hip_center - self.obs[58:61]) * 10
|
|
self.obs[58:61] = ball_rel_hip_center
|
|
self.obs[61] = np.linalg.norm(ball_rel_hip_center) * 2
|
|
self.obs[62] = U.normalize_deg(self.kick_ori - r.imu_torso_orientation) / 30
|
|
'''
|
|
Expected observations for walking parameters/state (example):
|
|
Time step R 0 1 2 0 1 2 3 4
|
|
Progress 1 0 .5 1 0 .25 .5 .75 1
|
|
Left leg active T F F F T T T T T
|
|
Parameters A A A B B B B B C
|
|
Example note: (A) has a step duration of 3ts, (B) has a step duration of 5ts
|
|
'''
|
|
return self.obs
|
|
|
|
def sync(self):
|
|
''' Run a single simulation step '''
|
|
r = self.player.world.robot
|
|
self.player.scom.commit_and_send(r.get_command())
|
|
self.player.scom.receive()
|
|
|
|
def reset(self):
|
|
# print("reset")
|
|
'''
|
|
Reset and stabilize the robot
|
|
Note: for some behaviors it would be better to reduce stabilization or add noise
|
|
'''
|
|
self.lock_flag = False
|
|
self.player.scom.unofficial_set_play_mode("PlayOn")
|
|
Gen_ball_pos = [random() * 5 - 9, random() * 6 - 3, 0]
|
|
Gen_player_pos = (random() * 3 + Gen_ball_pos[0], random() * 3 + Gen_ball_pos[1], 0.5)
|
|
Gen_player_rot = random() * 360
|
|
self.ball_pos = np.array(Gen_ball_pos)
|
|
self.player.scom.unofficial_move_ball((Gen_ball_pos[0], Gen_ball_pos[1], Gen_ball_pos[2]))
|
|
self.sleep = 0
|
|
self.step_counter = 0
|
|
self.behavior = self.player.behavior
|
|
r = self.player.world.robot
|
|
w = self.player.world
|
|
t = w.time_local_ms
|
|
self.path_manager = self.player.path_manager
|
|
gait: Step_Generator = self.behavior.get_custom_behavior_object("Walk").env.step_generator
|
|
self.reset_time = t
|
|
|
|
for _ in range(25):
|
|
self.player.scom.unofficial_beam(Gen_player_pos, Gen_player_rot) # beam player continuously (floating above ground)
|
|
self.player.behavior.execute("Zero_Bent_Knees")
|
|
self.sync()
|
|
|
|
# beam player to ground
|
|
self.player.scom.unofficial_beam(Gen_player_pos, Gen_player_rot)
|
|
r.joints_target_speed[
|
|
0] = 0.01 # move head to trigger physics update (rcssserver3d bug when no joint is moving)
|
|
self.sync()
|
|
|
|
# stabilize on ground
|
|
for _ in range(7):
|
|
self.player.behavior.execute("Zero_Bent_Knees")
|
|
self.sync()
|
|
delta_diff = gauss(1, 0.1)
|
|
# walk to ball
|
|
while True and w.time_local_ms - self.reset_time <= 50000:
|
|
direction = 0
|
|
if self.player.behavior.is_ready("Get_Up"):
|
|
self.player.behavior.execute_to_completion("Get_Up")
|
|
self.bias_dir = [0.09, 0.1, 0.14, 0.08, 0.05][r.type]
|
|
biased_dir = M.normalize_deg(direction + self.bias_dir) # add bias to rectify direction
|
|
ang_diff = abs(
|
|
M.normalize_deg(biased_dir - r.loc_torso_orientation)) # the reset was learned with loc, not IMU
|
|
|
|
next_pos, next_ori, dist_to_final_target = self.path_manager.get_path_to_ball(
|
|
x_ori=biased_dir, x_dev=-self.ball_x_center, y_dev=-self.ball_y_center, torso_ori=biased_dir)
|
|
if (w.ball_last_seen > t - w.VISUALSTEP_MS and ang_diff < 5 + delta_diff and
|
|
t - w.ball_abs_pos_last_update < 100 and # ball absolute location is recent
|
|
dist_to_final_target < 0.018 and # if absolute ball position is updated
|
|
not gait.state_is_left_active and gait.state_current_ts == 2): # to avoid kicking immediately without preparation & stability
|
|
break
|
|
else:
|
|
dist = max(0.07, dist_to_final_target)
|
|
reset_walk = self.behavior.previous_behavior != "Walk" # reset walk if it wasn't the previous behavior
|
|
self.behavior.execute_sub_behavior("Walk", reset_walk, next_pos, True, next_ori, True,
|
|
dist) # target, is_target_abs, ori, is_ori_abs, distance
|
|
|
|
self.sync()
|
|
|
|
# memory variables
|
|
self.lastx = r.cheat_abs_pos[0]
|
|
self.act = np.zeros(self.no_of_actions, np.float32)
|
|
|
|
return self.observe(True)
|
|
|
|
def render(self, mode='human', close=False):
|
|
return
|
|
|
|
def close(self):
|
|
Draw.clear_all()
|
|
self.player.terminate()
|
|
|
|
def step(self, action):
|
|
r = self.player.world.robot
|
|
b = self.player.world.ball_abs_pos
|
|
w = self.player.world
|
|
t = w.time_local_ms
|
|
if self.step_counter < 6:
|
|
r.joints_target_speed[2:18] = action * [
|
|
2.1,
|
|
2.1,
|
|
1.1,
|
|
1.1,
|
|
0.21,
|
|
0.21,
|
|
0.21,
|
|
0.21,
|
|
2.2,
|
|
2.2,
|
|
1.1,
|
|
1.1,
|
|
1.1,
|
|
1.1,
|
|
1.1,
|
|
1.1]
|
|
r.joints_target_speed[12] += 1
|
|
r.joints_target_speed[6] += 3
|
|
r.joints_target_speed[5] -= 1
|
|
r.joints_target_speed[8] += 3
|
|
r.joints_target_speed[9] -= 5
|
|
else:
|
|
r.joints_target_speed[2:18] = action * [
|
|
7.03,
|
|
7.03,
|
|
3,
|
|
3,
|
|
7.03,
|
|
7.03,
|
|
7,
|
|
7,
|
|
7,
|
|
7,
|
|
2,
|
|
2,
|
|
1,
|
|
1,
|
|
1,
|
|
1]
|
|
r.joints_target_speed[6] -= 2.5
|
|
r.joints_target_speed[7] += 2.5
|
|
r.joints_target_speed[9] += 2.5
|
|
r.joints_target_speed[11] -= 0.7
|
|
r.set_joints_target_position_direct([0, 1], np.array([0, -44], float), False)
|
|
|
|
self.sync() # run simulation step
|
|
self.step_counter += 1
|
|
self.lastx = r.cheat_abs_pos[0]
|
|
|
|
# terminal state: the robot is falling or timeout
|
|
if self.step_counter > 16:
|
|
obs = self.observe()
|
|
self.player.scom.unofficial_beam((-14.5, 0, 0.51), 0) # beam player continuously (floating above ground)
|
|
waiting_steps = 0
|
|
high = 0
|
|
while waiting_steps < 650: # 假设额外等待5个步骤
|
|
if w.ball_cheat_abs_pos[2] > high:
|
|
high = w.ball_cheat_abs_pos[2]
|
|
self.sync() # 继续执行仿真步骤
|
|
waiting_steps += 1
|
|
dis = np.linalg.norm(self.ball_pos - w.ball_cheat_abs_pos)
|
|
reward = dis - abs(w.ball_cheat_abs_pos[1] - self.ball_pos[1]) + high*0.2
|
|
# print(reward)
|
|
self.terminal = True
|
|
|
|
else:
|
|
obs = self.observe()
|
|
reward = 0
|
|
self.terminal = False
|
|
|
|
return obs, reward, self.terminal, {}
|
|
|
|
|
|
class Train(Train_Base):
|
|
def __init__(self, script) -> None:
|
|
super().__init__(script)
|
|
|
|
def train(self, args):
|
|
|
|
# --------------------------------------- Learning parameters
|
|
n_envs = min(14, os.cpu_count())
|
|
n_steps_per_env = 128 # RolloutBuffer is of size (n_steps_per_env * n_envs)
|
|
minibatch_size = 64 # should be a factor of (n_steps_per_env * n_envs)
|
|
total_steps = 30000000
|
|
learning_rate = 3e-5
|
|
folder_name = f'Kick_15M_R{self.robot_type}'
|
|
model_path = f'./scripts/gyms/logs/{folder_name}/'
|
|
|
|
# print("Model path:", model_path)
|
|
|
|
# --------------------------------------- Run algorithm
|
|
def init_env(i_env):
|
|
def thunk():
|
|
return Kick_15M(self.ip, self.server_p + i_env, self.monitor_p_1000 + i_env, self.robot_type, False)
|
|
|
|
return thunk
|
|
|
|
servers = Server(self.server_p, self.monitor_p_1000, n_envs + 1) # include 1 extra server for testing
|
|
|
|
env = SubprocVecEnv([init_env(i) for i in range(n_envs)])
|
|
eval_env = SubprocVecEnv([init_env(n_envs)])
|
|
|
|
try:
|
|
if "model_file" in args: # retrain
|
|
model = PPO.load(args["model_file"], env=env, device="cuda", n_envs=n_envs, n_steps=n_steps_per_env,
|
|
batch_size=minibatch_size, learning_rate=learning_rate)
|
|
else: # train new model
|
|
model = PPO("MlpPolicy", env=env, verbose=1, n_steps=n_steps_per_env, batch_size=minibatch_size,
|
|
learning_rate=learning_rate, device="cuda")
|
|
|
|
model_path = self.learn_model(model, total_steps, model_path, eval_env=eval_env,
|
|
eval_freq=n_steps_per_env * 20, save_freq=n_steps_per_env * 20,
|
|
backup_env_file=__file__)
|
|
except KeyboardInterrupt:
|
|
sleep(1) # wait for child processes
|
|
print("\nctrl+c pressed, aborting...\n")
|
|
servers.kill()
|
|
return
|
|
|
|
env.close()
|
|
eval_env.close()
|
|
servers.kill()
|
|
|
|
def test(self, args):
|
|
# Uses different server and monitor ports
|
|
server = Server(self.server_p - 1, self.monitor_p, 1)
|
|
env = Kick_15M(self.ip, self.server_p - 1, self.monitor_p, self.robot_type, True)
|
|
model = PPO.load(args["model_file"], env=env)
|
|
|
|
try:
|
|
self.export_model(args["model_file"], args["model_file"] + ".pkl",
|
|
False) # Export to pkl to create custom behavior
|
|
self.test_model(model, env, log_path=args["folder_dir"], model_path=args["folder_dir"])
|
|
except KeyboardInterrupt:
|
|
print()
|
|
|
|
env.close()
|
|
server.kill()
|